Comparison with Previous Results

These simulations suggest that the use of CS rather than C configuration would make imaging some sources more difficult, particularly for snapshot observations. Yet previous work comparing images made with the two configurations (Holdaway 1994, Rupen 1997) uniformly praised CS' imaging capability. How can these results be reconciled? Holdaway (1994) was interested in how well CS could recover total flux densities, and how well one could map very extended sources both with single pointings and with mosaics. He naturally concentrated on very extended objects, for which CS is indeed far superior. For Cas A (at a declination of $20^\circ$) at the size used here he found that CS outperformed C configuration (for a $\pm2\rm\,hr$ observation) by about 15%, as measured by the median fidelity. The simulations reported here give the identical result for the VTESS images, an agreement which must in part be fortuitous, since my simulation was of a source at $\delta=60^\circ$ observed for $\pm4\rm\,hrs$ with a significant contribution from the thermal noise. Rupen (1997) was also primarily concerned with the recovery of the total flux density, and with the behavior of CS configuration in the situation for which it was designed: imaging big, low SNR HI. Furthermore that was an observational rather than a computer-based study, using real data to show that CS did indeed outperform C in real life. It is impossible to compare those results directly with the simulations presented here, but it is not surprising that CS would again do a good job for what is was designed to do.

By contrast, this memorandum confronts CS with images which match the design goals of the original C configuration, and which are particularly susceptible to the trade-offs inherent in moving two C configuration antennas to the innermost part of the array. I have tried to make things as bad as seems plausible for CS configuration, dumping two antennas which are likely to exaggerate its most worrisome characteristics, simulating observations of sources which will not benefit from the improved short-spacing information, using CLEAN rather than MEM lest the latter's preference for smoothness emphasize that new information. While the earlier work showed that CS gives amazing improvements in imaging capability for large sources, the current simulations show virtually no difference between the two configurations for long observations, and at most a factor two difference for single snapshots. The surprise is not that CS does so poorly, but that in these trying circumstances it doesn't do much worse.